Do Structural Traits Support Reliable Tree Species Classification in Savannas ? Insights From (…)

ABSTRACT

Understanding whether structural traits reliably reflect species identity is crucial for advancing automated tree classification, particularly in complex disturbance-prone ecosystems like the West Sudanian savannas. In this study, we assessed the ecological consistency and taxonomic discriminability of 17 architectural traits, spanning crown, stem branch and whole-tree components, derived from close-range photogrammetry point clouds across five dominant savanna tree species. Using a trait-by-trait analysis grounded in the Trait Probability Density (TPD) framework, we quantified intraspecific versus interspecific variation through functional richness (FRic) and pairwise trait overlap. We computed a composite score combining these two indicators to rank traits according to their species-discriminatory potential. ANOVA and the Student Newman Keuls post hoc test were performed on the composite score to identify groups of significantly distinct traits. Our findings showed that while structural traits exhibit considerable within-species variability, a subset of traits, particularly crown evenness, stem branch angle and stem taper factor, demonstrated relatively high interspecific structuring and low trait overlap. Crown and stem branch traits emerged as the most reliable for distinguishing species. However, the top performing traits were distributed across all architectural domains, underscoring the value of a multi-trait approach. We also observed species-specific inconsistencies even among high-ranking traits, emphasising the need to account for intraspecific plasticity in classification tasks. These results suggest that no single architectural trait offers universal taxonomic resolution, and call for further research into identifying context-specific structural features, particularly those robust to environmental disturbance and ecotypic variation, for advancing automated tree species classification in savanna systems.